4 Comments
User's avatar
Michael Thompson's avatar

Love this analogy. Thanks for the TL;DR summary. Breaks silently…and a user just leaves. Yep. Same as data. Except: in the worst possible moment they bring it up in a public forum 🫠

Alejandro Aboy's avatar

You can have the best AI Agent until you show it in a demo 🤣

Pawel Jozefiak's avatar

The boundary question is interesting - at what point does 'an agent that processes data' become a data product vs. just an automation? I'd argue the threshold is SLA: when you start treating the agent's outputs as something other teams depend on, with expectations around freshness, accuracy, and uptime, it's a data product.

The RAG + PGVector setup you mentioned from Alejandro is a good test case - if other teams are querying that output, it's a product whether or not you call it one. The ownership model that comes with that label is probably what matters most organizationally.

Alejandro Aboy's avatar

True Pawel, ownership on AI projects will be a super relevant discussion since right now most of them start as a side quest without too much definition, and that's where data & AI teams should work together mostly.